The present invention relates to a controller of a disk array apparatus which stores data on multiple hard disk drives.
As compared with the computer main storage which is based on the semiconductor memory, the disk subsystem (will be termed simply “subsystem”) which is based on the magnetic disk memory is sluggish in regard to the I/O performance by a degree of the order of 3–4 digits, and continuous efforts have been paid to reduce the difference of these memories, i.e., improve the I/O performance of the subsystem.
A known scheme to improve the I/O performance of the subsystem is the adoption of a disk array system in which multiple hard disk drives are organized to form a subsystem, and data is stored on the multiple hard disk drives.
For example, a prior art system shown in FIG. 2 includes multiple channel IF units 11 which implement the data transfer between host computers 50 and disk array controllers 2, multiple disk IF units 12 which implement the data transfer between magnetic disk drives 5 and disk array controllers 2, cache memory units 14 which temporarily store data of the hard disk drives 5, and shared memory units 13 which store control information for the disk array controllers 2 (e.g., information on data transfer control between the channel IF units 11 and disk IF units 12 and the cache memory units 14 and control information for data stored in the hard disk drives 5). In each disk array controller 2, the shared memory unit 13 and cache memory unit 14 can be accessed by all channel IF units 11 and disk IF units 12.
In the disk array controller 2, the channel IF units 11 and disk IF units 12 are connected to the shared memory unit 13 by an interconnection 21, and the channel IF units 11 and disk IF units 12 are connected to the cache memory unit 14 by another interconnection 22.
The channel IF unit 11 has the interface for the connection to host computers 50 and a microprocessor (not shown) which controls the data transaction with the host computers 50. The disk IF unit 12 has the interface for the connection to hard disk drives 5 and a microprocessor (not shown) which controls the data transaction with the hard disk drives 5. The disk IF units 12 have the RAID function.
This conventional disk array controller 2 has an upper limit of disk storage capacity which can be connected to it, and therefore for storing data in excess of the volume of data manageable by one disk array controller 2, multiple disk array controllers 2 are installed and channels from the host computers 50 are connected to the disk array controllers 2. Furthermore, for connecting host computers 50 in excess of the number of host channels which can be connected to one disk array controller 2, multiple disk array controllers 2 are installed and connected to the host computers 50 individually.
For data transfer between two disk array controllers 2, channels from the host computers 50 are connected to the two disk array controllers 2 and data is transferred via the host computers 50.
Another prior art system shown in FIG. 3, which is disclosed in U.S. Pat. No. 5,680,640, is designed for example to transfer data between two disk array controllers 3 based on the connection of part (two lines in FIG. 3) of the computer interface paths of the disk array controllers 3 and the data transfer from a hard disk drive 5 which is connected to one disk array controller 3 to another hard disk drive 5 which is connected to another disk array controller 3 through the data transfer path 8.
Another prior art system shown in FIG. 4 is designed for example to store data in excess of the volume of data manageable by one disk array controller, connect host computers in excess of the number of host channels which can be connected to one disk array controller, or transfer data among multiple disk array controllers, as shown in FIG. 4, based on the installation of multiple disk array controllers 4 and the connection of their computer interface paths to the host computers 50 via a switch-based interconnection 23.
A disk array processing device disclosed in Japanese Unexamined Patent Publication No.Hei-11-66693 includes two director units which use a shared memory to recover data when a data spindle which forms a disk array runs out of control. This patent publication, however, does not show the installation of multiple disk arrays.
In large business enterprises which typically include banks, stock traders, and telephone companies, there are trends of cutting the expenditure for the operation, maintenance and management of computer systems and storage systems based on the centralized installation of computers and storages which have been installed distributively in many places. In this movement with the times, large high-end disk array controllers are required to bear the channel interface for the connection to several-hundreds or more host computers and huge storage capacities of several-hundreds terabytes or more.
At the same time, due to growing open markets in recent years and storage area networks (SANs) which are expected to prevail in coming years, there are rising demands of small-scale (compact) disk array controllers which are comparable in performance and reliability with large high-end disk array controllers.
The former requirement will conceivably be met by connecting multiple conventional large high-end disk array controllers to build a huge disk array controller. The latter requirement will conceivably be met by arranging a lowest-level model (e.g., with small numbers of channel IF units and disk IF units) of a conventional large high-end disk array controller. An additional conceivable scheme is to connect a plurality of this compact disk array controller thereby to build a controller which supports systems of medium to large scales which have been supported by a conventional disk array controller.
Accordingly, it becomes necessary for disk array controllers to have the scalability so as to be constructed to cover the range from a small to huge-scale controllers based on the same basic architecture of high performance and high reliability, and therefore there arises a demand of a disk array controller, a plurality of which are integrated to operate as a single disk array controller.
However, although the prior art system shown in FIG. 2 can have an increased number of channels and increased storage capability for the host computers 50 by simply increasing the number of disk array controllers 2, the host computers 50 need to connect channels to all disk array controllers 2 so that one host computer 50 can transact data with multiple disk array controllers 2. Moreover, it is necessary for one host computer 50 which is going to make access to data to identify a disk array controller 2 that is connected to the hard disk drive 5 which stores the data. On this account, it is difficult to operate multiple disk array controllers as a single disk array controller.
The prior art system shown in FIG. 3 has disk array controllers 3 interconnected by the data transfer path 8, enabling one host computer 50 which is connected to a certain disk array controller 3 to make access to data on a hard disk drive 5 which is connected to other disk array controller 3, and accordingly it is possible to operate multiple disk array controllers 3 as a single disk array controller.
However, in case a host computer 50 issues a data read request to a disk array controller 3 and the data is absent on the hard disk drives 5 connected to the controller 3, it is necessary for the disk array controller 3 to send the read request to other disk array controller 3 over the data transfer path 8, receive the requested data over the data transfer path 8 from the disk array controller 3 which is connected to the hard disk drive 5 on which the data is stored, and send the requested data to the host computer 50. On this account, this system suffers a degraded performance when a host computer 50 makes access to data which is stored on a hard disk drive 5 connected to a disk array controller 3 other than the disk array controller 3 connected to itself.
For coping with this matter, if a host computer 50 transfers in advance data of frequent access, which is stored on a hard disk drive of a disk array controller other than that connected to itself, to the hard disk drive 5 of the disk array controller connected to itself, the system also suffers a degraded performance due to the data transfer over the data transfer path 8.
In the prior art system shown in FIG. 4, each host computer 50 can access to all disk array controllers 4 via the switch-based interconnection 23. However, in order to operate multiple disk array controllers 4 as a single disk array controller, each switch of the interconnection 23 must have a map indicating as to which of all disk array controllers 4 connected to the switch stores data, and must have a function of analyzing the access request from a host computer 50 and designating a disk array controller 4 which stores the requested data.
Due to the need of request command analysis by the switches besides the command analysis by the conventional channel IF units 11, this system suffers a degraded performance in contrast to the direct connection of the host computers 50 to the disk array controllers 4.
The high-end disk array controllers have following functions.
As a specific operational function, a data set (corresponding to a logical volume) for one work is duplicated and stored and both the original and duplicate data sets are updated simultaneously in the ordinary work. At the emergence of the need of backup for the data set for example, updating of the duplicate data set is suspended and it is used for the backup, while the original data set is used continuously for the work, and the original and duplicate data sets are rendered the matching process on completion of backup.
For accomplishing this operational function, with duplicate data sets being held in different disk array controllers, the prior art systems shown in FIG. 2 to FIG. 4 all need to transfer data sets among the disk array controllers, resulting in a significantly deteriorated system performance.